Re: [Discuss-gnuradio] inband pings and clock

[George Nychis]

Eric Blossom wrote:
> On Fri, Aug 03, 2007 at 11:12:30AM -0400, George Nychis wrote:
> > What is 'pingval' and what is its units?  I didn't think most pings had 
> > a value, only empty responses for which you could compute whatever delay 
> > value you please by calculating the time between.
> > http://gnuradio.org/trac/browser/gnuradio/branches/developers/gnychis/inband/usrp/host/lib/inband/usrp_server.mbh#L105
>
> The pingval is arbitrary.

Check!

> > Second question.  For an application to properly timestamp outgoing 
> > packets, the application needs some general idea of the current clock 
> > value on the USRP.  At first I was thinking "oh well the app can just 
> > send a ping and read the timestamp off of the response while calculating 
> > the delay."  Well, the ping doesn't carry the clock back up to the 
> > application.  RX packets (response-recv-raw-samples) carry the timestamp 
> > back up to the application, but there is no notion of calculating delay 
> > here.
>
> The usrp_server should be passing the timestamp back to the
> application.  It's in the header of the packet containing the ping
> reply.  We should add the timestamp to the response-from-control-channel

Right, this is where the current hole is.  I will add a timestamp field 
for the response-from-control-channel.


> > Is the clock stored in a readable register somewhere on the USRP that a 
> > register read could be used?  The delay could be calculated here between 
> > the request and response.  I checked our wiki but see no register:
> > http://gnuradio.org/trac/wiki/UsrpFPGA
>
> I think having the clock counter mapped to a register is a fine idea.
> Note that the list of regs at http://gnuradio.org/trac/wiki/UsrpFPGA
> is the existing list of regs, and hasn't been refactored to reflect
> the requirements for inband signaling.  Also, I think it's a bad idea
> to use the wiki for data where the ultimate source is the source code.
> A link to the source would make more sense, otherwise the wiki is
> pretty much guaranteed to be out-of-date.  (Basically we're violating
> the "write it once" principle)

Got ya, I'll stick with the source code.  We'll work out mapping the 
clock to a register.

- George